Multi-zone load prediction method and system fusing global enhancement and local attention characteristics

Abstract

The invention discloses a multi-region load prediction method and system fusing global enhancement and local attention features. The method comprises the following steps: establishing a load feature extraction model to extract load features from load data of different regions; establishing a meteorological feature extraction model to extract meteorological features from the multivariate meteorological data of different regions; establishing a multi-region load prediction model, wherein the multi-region load prediction model comprises a multi-scale LSTM feature extraction network formed by a plurality of LSTM models with different scales, a multi-layer perceptron network and a multi-region load prediction network established based on a transformer model; the multi-scale LSTM feature extraction network extracts load global features and load local features from the load features, and the multi-layer perceptron network fuses the load global features and the load local features with meteorological features to obtain an input data set of the multi-region load prediction network; and according to the input data set, the multi-region load prediction network outputs a multi-region load prediction result, so that the accuracy of the multi-region load prediction result is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of load forecasting, and in particular, relates to a multi-region load forecasting method and system integrating global enhancement and local attention features. Background Art

[0002] The reliability and stability of the power system are crucial to modern society, and load forecasting, as the basis for power system scheduling and planning, plays a vital role in achieving efficient energy utilization and providing stable power supply. With the growth of energy demand and the continuous development of new energy technologies, load forecasting has received increasing attention.

[0003] However, in the existing technology, traditional load forecasting methods usually focus on the load situation in a single region, ignoring the complex coupling characteristics between loads and meteorology in multiple regions, as well as the correlation between loads in multiple regions. At the same time, the existing technology does not conduct multi-regional interactive forecasting, resulting in the inability to fully consider the mutual influence between different regions, thereby limiting the load forecasting model's comprehensive capture of the dynamic changes of the overall power system, thereby reducing the accuracy and robustness of load forecasting. This deficiency is manifested in the difficulty of the model to provide accurate forecasting results when dealing with fluctuations in power demand among multiple regions, thus affecting the planning and scheduling decisions of the power system.

[0004] In the prior art, lightning sensitive loads in multiple regions are mined, and sample set data of all lightning sensitive loads in the regions to be mined are numbered; the long-term growth trend change component in the daily load curve of the current region to be mined is stripped; the fitting result function of the daily load curve after stripping the long-term growth trend change component of the load curve is obtained by fitting the meteorological conditions and date type identification information; the lightning sensitive load of the region on the day is calculated based on the average value of the difference between the predicted load and the actual load within the lightning impact time interval; the lightning sensitive loads in all regions are regionally aggregated to obtain all lightning sensitive load values ​​of all lightning sensitive loads in the regions to be mined, and the fitting of the relevant external meteorological and date information to the target daily load curve provides a reference for analyzing the characteristics of user electricity consumption behavior, but the accuracy of load prediction based on the fitting result is low.

[0005] Therefore, in-depth research on multi-regional load forecasting and exploring its mutual impact with multi-regional meteorology are of great significance for improving the accuracy and reliability of load forecasting. Summary of the invention

[0006] In order to solve the deficiencies in the prior art, the present invention provides a multi-region load forecasting method and system integrating global enhancement and local attention features, which improves the accuracy of multi-region load forecasting results by fully mining the load data characteristics under different regional conditions.

[0007] The present invention adopts the following technical solution.

[0008] The present invention proposes a multi-region load forecasting method integrating global enhancement and local attention features, comprising:

[0009] Obtain load data and multivariate meteorological data for multiple regions;

[0010] Based on the long short-term memory network and spatial attention mechanism, a load feature extraction model is established to extract load features from load data in different regions. Based on the token to token model, Vision Transformer model and Transformer model, a meteorological feature extraction model is established to extract meteorological features from multivariate meteorological data in different regions.

[0011] A multi-region load forecasting model is established, including: a multi-scale LSTM feature extraction network composed of multiple LSTM models of different scales, a multi-layer perceptron network, and a multi-region load forecasting network based on a transformer model; wherein the multi-scale LSTM feature extraction network extracts global load features and local load features from load features, and the multi-layer perceptron network fuses the global load features and local load features with meteorological features to obtain an input data set of the multi-region load forecasting network; based on the input data set, the multi-region load forecasting network outputs a multi-region load forecasting result.

[0012] Preferably, the multivariate meteorological data includes: wind speed, wind direction, wind force, humidity, atmospheric pressure, and temperature.

[0013] Preferably, the load data of multiple regions and multivariate meteorological data are processed, including:

[0014] Remove outliers, fill in missing values, and perform time series smoothing on the data set after removing outliers and filling in missing values;

[0015] Among them, the moving average method is used to perform time series smoothing on the data set after deleting outliers and filling missing values.

[0016] Preferably, the load feature extraction model includes: an input layer, N spatial attention modules, N long short-term memory network modules, and an output layer;

[0017] The multi-region load data is processed by the first spatial attention module and input into the first long short-term memory network module. The output data of the first long short-term memory network module is spliced ​​with the multi-region load data and input into the second spatial attention module. The data processed by the second spatial attention module and the output data of the first long short-term memory network module are input into the second long short-term memory network module together. By analogy, the output data of the N-1th long short-term memory network module is spliced ​​with the multi-region load data and input into the Nth spatial attention module. The data processed by the Nth spatial attention module and the output data of the N-1th long short-term memory network module are input into the Nth long short-term memory network module together. The data output by the Nth long short-term memory network module is the multi-region load feature. The multi-region load feature output by the Nth long short-term memory network module is transmitted to the output layer.

[0018] Preferably, each spatial attention module has the same structure, and includes: a load data input unit, a linear mapping layer, and an activation function unit;

[0019] The multi-regional load data are input into the linear mapping layer after passing through the load data input unit. The linear mapping layer includes three linear mapping units. The multi-regional load data are respectively obtained through the three linear mapping units to obtain three weighted matrices, which satisfy the following relationship:

[0020] Q=W Q ·X

[0021] K=W K ·X

[0022] V=W V ·X

[0023] Where Q is the first weight matrix W output by the first linear mapping unit Q The weighted sum matrix of the multi-region load data X is defined as the query matrix, and K is the second weight matrix W output by the second linear unit. K and the weighted sum matrix of multi-region load data X, defined as the key matrix, V m The third weight matrix W output by the third linear emitter unit V and the weighted sum matrix of multi-region load data X, defined as the value matrix;

[0024] The activation function unit Softmax obtains the output data of the mth spatial attention module according to the following relationship:

[0025]

[0026] Where Attention(Q, K, V) is the output data of the spatial attention module, dk is the scaling factor, and the value of the scaling factor dk is the number of columns of the query matrix Q and the key matrix K.

[0027] Preferably, the meteorological feature extraction model comprises: an input layer, N T2T modules, N Vision Transformer modules, N Transformer modules, and an output layer;

[0028] After passing through the first T2T module, the multivariate meteorological data is first embedded in a patch and then input into the first Vision Transformer module. The data output by the first Vision Transformer module is transmitted to the first Transformer module. After passing through the second T2T module together with the multivariate meteorological data, the data output by the first Transformer module is first embedded in a patch and then input into the second Vision Transformer module. The data output by the second Vision Transformer module and the data output by the first Transformer module are transmitted to the second Transformer module together. Similarly, after passing through the Nth T2T module together with the multivariate meteorological data, the data output by the N-1th Transformer module is first embedded in a patch and then input into the Nth Vision Transformer module. The data output by the Nth Vision Transformer module and the data output by the N-1th Transformer module are transmitted to the Nth Transformer module together. The meteorological features output by Transformer block N of the Nth Transformer module are transmitted to the output layer.

[0029] Preferably, each T2T module comprises: f processing units and f-1 conversion units;

[0030] In the conversion unit, the i-th element in the sequence is updated using the weighted sum of the i-th element in the sequence and other elements in the sequence, and a new sequence is obtained after all elements are updated;

[0031] Among them, the weights used in weighted summation satisfy the following relationship:

[0032] α[i,j]=exp(score[i,j]) / Sum_k(exp(score[i,j]))

[0033] Where α[i,j] is the weight between the i-th element and the j-th element, socre[i,j] is the correlation between the i-th element and the j-th element, and Sum_k() is the sum.

[0034] Preferably, the multi-scale LSTM feature extraction network extracts load global features and load local features from load features, including:

[0035] Process the load characteristics and meteorological characteristics into a confusion matrix form;

[0036] The multi-scale LSTM feature extraction network includes N LSTM models of different scales; the first LSTM model takes the load feature in the form of a confusion matrix as input and outputs the first local feature of the load; the second LSTM model takes the first local feature of the load as input and outputs the second local feature of the load; and so on, the Nth LSTM model takes the N-1th local feature of the load as input and outputs the Nth local feature and the global feature of the load.

[0037] Preferably, the multi-layer perceptron network fuses the global load features and the local load features with the meteorological features to obtain an input data set for the multi-region load forecasting network, including:

[0038] The multi-layer perceptron network includes N multi-layer perceptron layers;

[0039] In the first multi-layer perceptron layer, the first local feature of the load is fused to obtain the first local fusion feature, and the first local fusion feature matches the dimension of the meteorological feature in the form of a confusion matrix; in the second multi-layer perceptron layer, the second local feature of the load is fused to obtain the second local fusion feature, and the second local fusion feature matches the dimension of the meteorological feature in the form of a confusion matrix; and so on, in the Nth multi-layer perceptron layer, the Nth local feature of the load is fused to obtain the Nth local fusion feature, and the global feature of the load is fused to obtain the global fusion feature, and the Nth local fusion feature and the global fusion feature both match the dimension of the meteorological feature in the form of a confusion matrix; the first local fusion feature, the second local fusion feature, ..., the Nth local fusion feature and the global fusion feature constitute the input data set of the multi-regional load forecasting network.

[0040] Preferably, based on the input data set, the multi-region load forecasting network outputs a multi-region load forecasting result, including:

[0041] The multi-region load forecasting network includes N transformer models. The meteorological feature input data is transmitted to the first transformer model together with the first local fusion feature. The output data of the first transformer model is transmitted to the second transformer model together with the second local fusion feature. And so on. The output data of the N-1th transformer model is transmitted to the Nth transformer model together with the Nth local fusion feature and the global fusion feature. The Nth transformer model outputs the multi-region load forecasting result.

[0042] The present invention also proposes a multi-region load forecasting system integrating global enhancement and local attention features, comprising:

[0043] The data processing module is used to obtain load data and multivariate meteorological data from multiple regions; a load feature extraction model is established based on the long short-term memory network and the spatial attention mechanism to extract load features from load data from different regions; a meteorological feature extraction model is established based on the token to token model, the Vision Transformer model, and the Transformer model to extract meteorological features from multivariate meteorological data from different regions;

[0044] The load forecasting module is used to establish a multi-region load forecasting model, including: a multi-scale LSTM feature extraction network composed of multiple LSTM models of different scales, a multi-layer perceptron network, and a multi-region load forecasting network established based on a transformer model; wherein the multi-scale LSTM feature extraction network extracts global load features and local load features from load features, and the multi-layer perceptron network fuses the global load features and local load features with meteorological features to obtain an input data set for the multi-region load forecasting network; based on the input data set, the multi-region load forecasting network outputs multi-region load forecasting results.

[0045] The present invention is also a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0046] The present invention is also a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0047] The beneficial effect of the present invention is that, compared with the prior art, at least the method proposed by the present invention extracts features from meteorological data using a NWP processing model based on T2T and Vision Transformer for numerical weather forecast (NWP) data such as temperature, humidity, wind speed, and atmospheric pressure. The global and local features of multi-regional loads extracted by a multi-layer perceptron MLP are fused with NWP features for input model training to construct a multi-regional load forecasting method that integrates global enhancement and local attention features, thereby overcoming the problems of lack of meteorological data and difficulty in joint forecasting of multi-regional loads and improving the accuracy of forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a multi-region load forecasting method integrating global enhancement and local attention features proposed by the present invention;

[0049] Figure 2 is a structural diagram of a load feature extraction model in an embodiment of the present invention;

[0050] Figure 3 It is a structural diagram of the NWP data feature extraction model in an embodiment of the present invention;

[0051] Figure 4 It is a flow chart of the MLP feature fusion model in an embodiment of the present invention;

[0052] Figure 5 It is a comparison diagram between the model load test curve and the actual value in the embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0054] The present invention proposes a multi-region load forecasting method integrating global enhancement and local attention features, comprising:

[0055] Step 1: Obtain load data and multivariate meteorological data from multiple regions.

[0056] In the embodiment, Figure 1 As shown, the input layer includes the load data of region 1, the load data of region 2, ..., the symbol data of region M, and also includes the multivariate meteorological data of region 1, the multivariate meteorological data of region 2, ..., the multivariate meteorological data of region M; the multivariate meteorological data comes from NWP; where M is the number of regions;

[0057] Specifically, the load data and NWP data of multiple regions are preprocessed to obtain the initial input data, including:

[0058] 1) Remove outliers:

[0059] There may be outliers in the load data and NWP data, so the outliers are deleted after outlier detection is performed on the collected data; in the embodiment, the method of outlier detection includes but is not limited to: 3σ method; specifically as follows: the load data and NWP data of each region constitute a data set, and the mean μ and standard deviation σ of the data set are calculated; three times the standard deviation 3σ is selected as the threshold; when the data sample x satisfies |x-μ|>3σ, the data sample x is determined to be an outlier.

[0060] The calculation formulas for the mean μ and standard deviation σ are as follows:

[0061]

[0062] In the formula, x i is the i-th data sample in the data set, and n is the total number of data samples.

[0063] 2) Filling missing values:

[0064] The missing values ​​are filled using the mean method as follows: the mean μ′ of the non-missing data of the data set is calculated, and all missing values ​​are filled using the mean μ′ of the non-missing data.

[0065] The calculation formula of the non-missing data mean μ′ is as follows:

[0066]

[0067] In the formula, n′ is the total number of non-missing data samples, x j is the jth non-missing data sample.

[0068] 3) Perform time series smoothing on the data set after deleting outliers and filling missing values ​​to eliminate noise;

[0069] The moving average method is used to smooth the data set after deleting outliers and filling missing values ​​to eliminate noise, as follows:

[0070] Set the sliding window width k, where k is an odd number greater than 1. The sliding window size determines the degree of smoothing. The larger the sliding window, the more obvious the smoothing effect, but it may cause over-smoothing of the data.

[0071] The data samples are processed into a time series data sequence according to the time series relationship of the sampling moments; when the central data point of the sliding window is aligned with the sampling moment t, the average value of the data in the sliding window is calculated as the moving average value at the sampling moment t; wherein, when the sliding window is at the head or tail of the time series data sequence, if the number of data samples in the sliding window is less than k, the moving average value is calculated based on the actual existing data samples and the number of data samples;

[0072] The calculation formula of the moving average value at sampling time t is as follows:

[0073]

[0074] In the formula, MA t is the moving average value at sampling time t, x i is the i-th data sample in the sliding window;

[0075] When the number of data samples in the sliding window is less than k, the calculation formula for the moving average at sampling time t is as follows:

[0076]

[0077] Where k1 and k2 are the number of data samples before and after the sampling time t in the sliding window, respectively, and k1+k2+1≤k;

[0078] The moving average of each sampling moment in the time series data sequence is calculated in turn, and the data is smoothed and noise is removed by reflecting the long-term trend.

[0079] Step 2: Based on the long short-term memory network (LSTM) and spatial attention mechanism, a load feature extraction model is established to extract load features from load data in different regions; based on the token to token model, VisionTransformer model and Transformer model, a meteorological feature extraction model is established to extract meteorological features from multivariate meteorological data in different regions.

[0080] Figure 1 In the processing layer, the load feature extraction model is established by using the space attention module Space attention, the hidden layer hidden layer, and the long short-term memory network modules LSTM Block1, LSTM Block1, ..., LSTM BlockN, and the T2T module and the Vision Transformer modules Transformer Block1, Transformer Block2, ..., Transformer Block N; wherein each long short-term memory network module LSTMBlock1, LSTM Block1, ..., LSTM Block n is connected in series in sequence, and the output of LSTM Block n is the output of the load feature extraction model; each Vision Transformer module Transformer Block1, TransformerBlock2, ..., Transformer Block N is connected in series in sequence, and the output of Transformer Block N is the output of the meteorological feature extraction model; wherein the number of the space attention module Space attention, the hidden layer hidden layer, the long short-term memory network module, the T2T module, and the Vision Transformer module is N;

[0081] Specifically, step 2 includes:

[0082] Step 2.1, based on the long short-term memory network (LSTM) and spatial attention mechanism, a load feature extraction model is established to extract load features from load data in different regions;

[0083] The specific model structure diagram is as follows Figure 2 As shown, the load data of region 1, region 2, region 3, ..., region M in the input layer are respectively input into the corresponding spatial attention layer; N spatial attention layers and N long short-term memory network modules LSTM block constitute a high-dimensional feature extraction layer; the multi-region load data are processed by the first spatial attention layer and then input into the first long short-term memory network module LSTM Block1, the output data of the first long short-term memory network module is spliced ​​with the multi-region load data and then input into the second spatial attention layer, the data processed by the second spatial attention layer and the output data of the first long short-term memory network module are input into the second long short-term memory network module together, and so on, the output data of the N-1th long short-term memory network module is spliced ​​with the multi-region load data and then input into the Nth spatial attention layer, the data processed by the Nth spatial attention layer and the output data of the N-1th long short-term memory network module are input into the Nth long short-term memory network module together; the data output by the Nth long short-term memory network module is the multi-region load feature data;

[0084] In the present invention, the output data of each spatial attention module includes: the spatial dependency of multi-region load data and the dynamic weight of each region load data; the spatial dependency of multi-region load data is a local attention feature, each spatial attention module effectively models the potential association between different regions, and can dynamically allocate the weight of different region loads, highlighting the contribution of important regions, while suppressing irrelevant or redundant information;

[0085] Each LSTM network module is used to combine the time dependency and spatial dependency of multi-regional load data; at the same time, the output data of the n-1th LSTM network module is the input data of the nth LSTM network module. Through the series connection of multiple LSTM modules, each module not only retains the time dependency information, but also gradually integrates the multi-regional global features extracted by the spatial attention module; the output data of the n-1th LSTM network module and the data after splicing the multi-regional load data are obtained in the nth spatial attention module. The global enhanced features are combined with the output data of the n-1th LSTM network module and input into the nth LSTM network module. This design, through the iterative update of each layer, allows the load characteristics of each region to not only integrate the time characteristics of the region, but also comprehensively consider the global impact of other regional characteristics.

[0086] The multi-regional load characteristics are used as input data of the high-dimensional feature fusion layer, which includes multiple fusion modules Merge-Model. The load feature extraction model also includes an output layer Output layer. The data output by the high-dimensional feature fusion layer is transmitted to the output layer Output layer, namely, the load feature Output-L.

[0087] The structure of each spatial attention module is the same, including: load data input unit Load data input, linear mapping layer Linear Layer, activation function unit Softmax; the spatial attention module performs weighted summation on the load data of different regions to mine the local and global features of the load in multiple regions with enhanced spatial correlation; the processing process of the mth spatial attention module on the load data of multiple regions is as follows:

[0088] The multi-regional load data is input into the linear mapping layer through the load data input unit. The linear mapping layer includes three linear mapping units. The multi-regional load data is respectively obtained through the three linear mapping units to obtain three weighted matrices, which satisfy the following relationship:

[0089] Q=W Q ·X

[0090] K=W K ·X

[0091] V=W V ·X

[0092] Where Q is the first weight matrix W output by the first linear mapping unit Q The weighted sum matrix of the multi-region load data X is defined as the query matrix, and K is the second weight matrix W output by the second linear unit. K and the weighted sum matrix of multi-region load data X, defined as the key matrix, V m The third weight matrix W output by the third linear emitter unit V and the weighted sum matrix of multi-region load data X, defined as the value matrix;

[0093] The activation function unit Softmax obtains the output data of the mth spatial attention module according to the following relationship:

[0094]

[0095] Where Attention(Q,K,V) is the output data of the spatial attention module, and dk is the scaling factor;

[0096] QK T is the similarity matrix between the query matrix Q and the key matrix K. In order to prevent the similarity from being too large, the similarity matrix QK T The ratio to the scaling factor dk is normalized by the Softmax function to convert the similarity matrix into an attention weight matrix, and the weighted sum of the attention weight matrix and the value matrix V is used as the output data of the spatial attention module; wherein the value of the scaling factor dk is the number of columns of the query matrix Q and the key matrix K.

[0097] In the spatial domain, the spatial attention mechanism is essentially to re-encode a matrix into a matrix after considering global information through matrix operations, while considering the global and local characteristics of load data in multiple regions, thereby strengthening the spatial correlation of the data. At the same time, LSTM, as a recurrent neural network, performs feature mining on the historical load data of each region and captures the changing patterns of load at different time scales. The core idea of ​​LSTM is to control the flow of information by introducing gating mechanisms, including forget gates, input gates, and output gates. These gating mechanisms allow LSTM networks to effectively capture long-term dependencies, making them suitable for processing time series data of loads.

[0098] The load feature extraction model established by the present invention is a high-dimensional feature extraction module, which contains N spatial attention and LSTM networks. Each network performs spatial and temporal feature extraction operations to mine local and global features of multi-regional load data and more comprehensively capture information at different time steps. The extracted high-dimensional feature sequence is used as the input of the spatial attention and LSTM network of the next time step, and so on, and the output at N sampling points is finally calculated. The multi-regional load high-dimensional features extracted by the model pay more attention to the chronological correlation of time series data, ensure a more comprehensive understanding and extraction of time series features in the data, and enable the model to better adapt to complex regional load data patterns.

[0099] Step 2.2, a meteorological feature extraction model is established based on the token to token model, the Vision Transformer model, and the Transformer model to extract meteorological features from multivariate meteorological data of different regions;

[0100] Multivariate meteorological data include but are not limited to: wind speed, wind direction, wind force, humidity, barometric pressure, and temperature;

[0101] The specific model structure diagram is as follows Figure 3 As shown in the figure, the multivariate meteorological data in the input layer are simultaneously input into the model; the multivariate meteorological data is first embedded into patches after passing through the first T2T module token to token module and then input into the first Vision Transformer module, and the data output by the first Vision Transformer module is transmitted to the first Transformer module Transformer block 1; the data output by the first Transformer module and the multivariate meteorological data are first embedded into patches after passing through the second T2T module and then input into the second Vision Transformer module, and the data output by the second Vision Transformer module and the data output by the first Transformer module are transmitted to the second Transformer module Transformer block 2 together; and so on, the data output by the N-1th Transformer module and the multivariate meteorological data are first embedded into patches after passing through the Nth T2T module and then input into the Nth Vision Transformer module, and the data output by the Nth Vision Transformer module and the data output by the N-1th Transformer module are transmitted to the Nth Transformer module Transformer block together. N; the data output by the Nth Transformer block Transformer block N is transmitted to the output layer Outputlayer, that is, the meteorological feature OutputT.

[0102] In order to fully explore the hidden features in NWP data from multiple regions, this section builds a meteorological data feature processing model based on token-to-token and VisionTransformer to extract features from the multivariate meteorological data collected from multiple regions.

[0103] Each T2T module consists of: f processing units Token 1, Token 2, ..., Token f and f-1 conversion units T2T transformer; the Tokens-to-Token module is designed to overcome the limitations of simple Tokenization in ViT. It gradually structures the image into tokens and models the local structural information, thereby gradually reducing the length of the token. Each T2T process has two steps: reorganization and soft splitting. Given a T sequence from the previous Transformer layer, it will be transformed by the self-attention block to satisfy the following relationship:

[0104] T′=MLP(MSA(T))

[0105] In the formula, T′ is the transformed sequence, T is the sequence, MSA() is the multi-head self-attention operation of multi-layer normalization, and MLP() is the multi-layer self-attention operation;

[0106] Reshape T′ into an image in the spatial dimension to satisfy the following relationship:

[0107] I=Re shape(T')

[0108] Where Reshape() is the image reshaping operation, and I is the reconstructed image of T′ in the spatial dimension;

[0109] After obtaining the reconstructed image I, it is soft-segmented to model the local structural information and reduce the length of the token. Specifically, in order to avoid information loss when generating tokens from the reconstructed image, it is segmented into image blocks with overlap. Therefore, each image block is related to the surrounding image blocks to establish a priori that there should be stronger correlations between the surrounding tokens. The tags in each segmented image block are connected as one tag, so that local information can be aggregated from the surrounding pixels and patches. When performing soft segmentation, the size of each image block is k×k, with s overlaps and p padding on the image, where ks is similar to the stride in the convolution operation. Therefore, for the reconstructed image i2r of size h×ω×c, the output token length after soft segmentation is:

[0110]

[0111] Where l0 is the output token length after soft segmentation;

[0112] The size of each segmentation patch is k×k×c; all image patches in the spatial dimension are flattened into tokens. After soft segmentation, the output tokens are provided to the next T2T process.

[0113] The T2T module iterates the above reconstruction and soft segmentation to gradually reduce the length of the token and transform the spatial structure of the image. The iterative process of the T2T module satisfies the following relationship:

[0114] T i =MLP(MSA(T i ))

[0115] I i =Reshape(T i ')

[0116] T i+1 =SS(I i),i=1....(n-1)

[0117] Where, T i ′ is the transformed sequence in the i-th iteration, T i is the sequence in the i-th iteration, I i is the reconstructed image in the i-th iteration, SS is the image transformation sequence operation, T i+1 is the sequence in the i+1th iteration, and n is the number of iterations;

[0118] For the input image I0, soft segmentation is first applied to segment it into tokens. After the final iteration, the output token T of the T2T module is n has a fixed length, so the trunk of T2T-vit can be n Modeling global relationships.

[0119] The Transformer structure is used inside the T2T model, and the core mechanism is the self-attention machine. In this process, each token will interact with other tokens in the sequence, and a weighted sum is performed based on the correlation between them to obtain a new token representation. The specific formula is as follows:

[0120] y[i]=Sum_j(α[i,j]*X[j])

[0121] Where y[i] is the i-th Token, X[j] is the j-th Token in the sequence, α[i,j] is the weight between the i-th Token and the j-th Token, and Sum_j() is the weighted sum. The weight is calculated as follows:

[0122] α[i,j]=exp(score[i,j]) / Sum_k(exp(score[i,j]))

[0123] In the formula, socre[i,j] is the correlation between the i-th Token and the j-th Token, and Sum_k() is the sum;

[0124] This process will be repeated in each layer of Transformer. After the new representation of all tokens in the final layer is obtained through the calculation of multiple layers of Transformer, an expansion operation is performed to map these representations to various meteorological scalars again. The new representation is an N-dimensional vector, which is re-expanded into the original token sequence of various dimensions such as temperature, humidity, and wind speed through a specific mapping relationship. Finally, based on these high-dimensional features, subsequent decoding operations are performed. This process not only ensures that the correlation between meteorological variables of various dimensions is fully considered, but also extracts richer features.

[0125] Step 3, establish a multi-region load forecasting model, including: a multi-scale LSTM feature extraction network composed of multiple LSTM models of different scales, a multi-layer perceptron network, and a multi-region load forecasting network established based on a transformer model; wherein the multi-scale LSTM feature extraction network extracts load global features and load local features from load features, and the multi-layer perceptron network fuses load global features and load local features with meteorological features to obtain an input data set of the multi-region load forecasting network; based on the input data set, the multi-region load forecasting network outputs a multi-region load forecasting result.

[0126] Specifically, step 3 includes:

[0127] Step 3.1, process the load characteristics and meteorological characteristics into a confusion matrix form;

[0128] In the embodiment, high-dimensional features extracted from load data and high-dimensional features extracted from meteorological data are reshaped into a shape suitable for a confusion model.

[0129] Step 3.2, a multi-scale LSTM feature extraction network is constructed using multiple LSTM models of different scales; according to the load characteristics in the form of confusion matrix, the multi-scale LSTM feature extraction network outputs the global characteristics and local characteristics of the load in multiple regions;

[0130] like Figure 4 As shown, the load feature input data in the form of confusion matrix is ​​transmitted to the LSTM out layer of the multi-scale LSTM feature extraction network, and the multi-scale LSTM feature extraction network extracts global features and local features from the load feature input data; wherein, the multi-scale LSTM feature extraction network includes N LSTM models of different scales, the load feature in the form of confusion matrix is ​​transmitted to LSTM block 1, LSTM block 1 outputs the first local feature of the load, the first local feature of the load is transmitted to LSTM block 2, LSTM block 2 outputs the second local feature of the load, and so on, LSTM block N outputs the Nth local feature of the load and the global feature of the load;

[0131] Step 3.3, using a multi-layer perceptron to perform dimension matching and feature fusion on the meteorological features in the form of confusion matrix and the global and local features of multi-regional loads, so that the local fusion features and global fusion features obtained after fusion constitute the input data set of the multi-regional load forecasting network;

[0132] like Figure 4As shown, the multilayer perceptron network Mixed layer includes N multilayer perceptron layers Multilayerperceptron layer, in the first multilayer perceptron layer, the first local feature of the load is subjected to feature fusion to obtain the first local fusion feature, and the first local fusion feature matches the dimension of the meteorological feature in the form of a confusion matrix; in the second multilayer perceptron layer, the second local feature of the load is subjected to feature fusion to obtain the second local fusion feature, and the second local fusion feature matches the dimension of the meteorological feature in the form of a confusion matrix; and so on, in the Nth multilayer perceptron layer, the Nth local feature of the load is subjected to feature fusion to obtain the Nth local fusion feature, and the global feature of the load is subjected to feature fusion to obtain the global fusion feature, and the Nth local fusion feature and the global fusion feature both match the dimension of the meteorological feature in the form of a confusion matrix; the first local fusion feature, the second local fusion feature, ..., the Nth local fusion feature and the global fusion feature constitute the input data set of the multi-region load forecasting network;

[0133] In the embodiment, the multi-layer perceptron MLP learns complex mapping relationships through nonlinear transformations at multiple levels, and can transform the dimensions of local and global features extracted from multi-regional loads, thereby matching the characteristics of meteorological data. First, the input layer accepts the original feature vector as input. Assume that the input layer has d neurons and the input vector is x = [x1, x2, ..., x d ], and secondly, because MLP contains one or more hidden layers, each hidden layer contains several neurons. Each neuron is connected to all neurons in the previous layer and has weights and biases. The weighted sum of the inputs of the ith neuron in the lth layer is as follows:

[0134]

[0135] In the formula, is the weighted sum of the inputs of the ith neuron in the lth input layer, d (l-1) is the number of neurons in the l-1th input layer, is the weight of the jth neuron in the l-1th input layer and the i-th neuron in the lth input layer, is the bias of the ith neuron in the lth input layer, is the output of the jth neuron in the l-1th input layer.

[0136] Then, a nonlinear transformation is performed through the activation function. The specific operation is shown in the following formula:

[0137]

[0138] In the formula, is the nonlinear transformation of the ith neuron in the lth input layer, and f() is the activation function;

[0139] Commonly used activation functions include sigmoid, tanh, ReLU, etc.

[0140] The output layer is usually used to generate the final prediction result. Similar to the hidden layer, each neuron in the output layer also has a weighted sum as shown below:

[0141]

[0142] In the formula, is the weighted sum of the kth neuron in the Lth output layer, d (L-1) is the number of neurons in the L-1th output layer, is the weight of the i-th neuron in the L-1th output layer and the k-th neuron in the Lth output layer, is the bias of the kth neuron in the Lth output layer, is the output of the i-th neuron in the L-1-th output layer.

[0143] The activation function of the output layer is for multi-classification problems, so the output result is obtained through the softmax function. Finally, the output value processed by MLP is merged with the last three layers in the transformer model. In the final transformer model, the meteorological characteristics and the global and local characteristics of the load in multiple regions are considered at the same time, which can effectively predict the load values ​​of multiple regions under different meteorological conditions.

[0144] The present invention uses meteorological characteristic data to promote the improvement of the reliability of multi-regional load forecasting results. The key lies in realizing the interaction of meteorological data characteristics and load data characteristics through the MLP feature fusion model. Including: the impact of meteorological characteristics on regional load is nonlinear and heterogeneous, and different regions have different sensitivities to meteorological conditions (such as temperature, humidity, wind speed); introducing meteorological data into the Transformer model and jointly modeling with the local fusion characteristics of the load. This design can capture the direct impact of meteorological conditions on the dynamic changes of load; global information sharing: through the joint input of global features and meteorological features, the impact of meteorological conditions on the overall load distribution of all regions is captured, thereby enhancing the synergy between regions.

[0145] Step 3.4: Establish a multi-region load forecasting network based on the transformer model.

[0146] like Figure 4As shown, the multi-region load forecasting network includes N transformer models. The meteorological feature input data is transmitted to the first transformer model Transformer block 1 together with the first local fusion feature. The data output by the first transformer model is transmitted to the second transformer model Transformer block 2 together with the second local fusion feature. By analogy, the data output by the N-1th transformer model is transmitted to the Nth transformer model Transformer block N together with the Nth local fusion feature and the global fusion feature. The Nth transformer model outputs the multi-region load forecasting result.

[0147] When the global and local features of multi-regional loads extracted by MLP are fused with NWP features and input into the Transformer model for training and prediction, an early stopping strategy is set. When the loss of the validation set does not improve within 30 training cycles (epochs), the training is stopped and the optimal model weight is restored. The model is trained using the data of the training set and verified using the data of the test set.

[0148] According to the multi-region load forecasting results, the reliability of the multi-region load forecasting model proposed in the present invention is evaluated using two commonly used model evaluation indicators, namely, root mean square error (RMSE) and mean absolute error (MAE).

[0149] The following three scenarios are set in the embodiment for comparative analysis:

[0150] 1) Scenario 1: Forecasting load when only globally enhanced feature fusion is performed; set the LSTM-out switch and use the LSTM model to output the load data separately. This output is directly connected to the input of the classifier. The output of the model directly depends on the output of the last time step of the LSTM to predict the load;

[0151] 2) Scenario 2: Load prediction without global enhancement and local attention feature fusion; set the MLP-Mixer switch, do not apply the MLP module after the meteorological and load feature extraction, so that the model does not convert and fuse the local and global features of meteorological and load, and then use the model for load prediction;

[0152] 3) Scenario 3: Load prediction when integrating global enhancement and local attention features; mining load and weather features through LSTM-Attention and Vit-Transformer, fusion processing of features through MLP, and prediction through Transformer.

[0153] By setting up the local attention feature extraction of LSTM-Attention and the MLP fusion feature module, the prediction model is subjected to an ablation experiment. The attention mechanism is not used, and a separate LSTM model is used to extract features from the load data. The model performance without local attention feature extraction is compared to evaluate the role of the local attention mechanism in the model. The MLP feature fusion mechanism is set to determine whether the MLP feature fusion module is applied after mining the global enhancement and local attention features, so that the model can perform deeper feature extraction and conversion on the features. This is to evaluate the performance difference of the model when using and not using the MLP feature fusion mechanism. The comparison curves of the prediction curves and actual values ​​of the three scenarios are shown in the figure below. Figure 5 As shown, Figure 5 The horizontal axis is the sampling time Time / Sample, the vertical axis is the model output value Y-value, the solid line is the predicted value Predicted value, and the dot-dashed line is the real value Real value; the load curve predicted by the model that mines the load and meteorological features of LSTM-Attention and Vit-Transformer, fuses the features through MLP, and then predicts through Transformer is closer to the real value. This is because this model combines the advantages of LSTM-Attention and Vision Transformer (ViT) and fuses the features through MLP. This architecture uses the LSTM-Attention model to capture long-term dependencies in time series data and focuses on key information through the attention mechanism. Then, the ViT model further extracts and processes the features of load and meteorological data, enhancing the model's ability to recognize complex patterns. After that, the MLP module transforms and fuses these features, improving the model's understanding of the relationship between features and the accuracy of prediction. Finally, the Transformer model uses its self-attention mechanism to effectively handle long-distance dependency problems and make accurate load forecasts. This multi-model fusion method not only improves the prediction accuracy and efficiency, but also enhances the model's adaptability and generalization ability to data changes. Therefore, the prediction ability of this model is better than that of the model in the ablation experiment.

[0154] Table 1 Evaluation of prediction results

[0155]

[0156] The comparison of prediction model evaluation parameters for the three scenarios is shown in Table 1. It can be seen that compared with the prediction models of scenarios 1 and 2, the multi-region load forecasting model that integrates global enhancement and local attention features can effectively improve the accuracy of load forecasting. The MAE values ​​of the load forecasting results of the method proposed in the present invention were reduced by 0.0051 and 0.0026 respectively; the RMSE was reduced by 0.0095 and 0.0094 respectively. Analysis of the above results shows that the MAE value and RMSE value of the load forecasting results have been reduced to a certain extent, which verifies that the model proposed in the present invention has good applicability in load forecasting.

[0157] The present invention also proposes a multi-region load forecasting system integrating global enhancement and local attention features, comprising:

[0158] The data processing module is used to obtain load data and multivariate meteorological data from multiple regions; a load feature extraction model is established based on the long short-term memory network and the spatial attention mechanism to extract load features from load data from different regions; a meteorological feature extraction model is established based on the token to token model, the Vision Transformer model, and the Transformer model to extract meteorological features from multivariate meteorological data from different regions;

[0159] The load forecasting module is used to establish a multi-region load forecasting model, including: a multi-scale LSTM feature extraction network composed of multiple LSTM models of different scales, a multi-layer perceptron network, and a multi-region load forecasting network established based on a transformer model; wherein the multi-scale LSTM feature extraction network extracts global load features and local load features from load features, and the multi-layer perceptron network fuses the global load features and local load features with meteorological features to obtain an input data set for the multi-region load forecasting network; based on the input data set, the multi-region load forecasting network outputs multi-region load forecasting results.

[0160] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0161] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0162] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0163] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A multi-region load forecasting method integrating global enhancement and local attention features, characterized in that: include: Obtain load data and multivariate meteorological data for multiple regions; Based on the long short-term memory network and spatial attention mechanism, a load feature extraction model is established to extract load features from load data in different regions; A meteorological feature extraction model is established based on the token to token model, the Vision Transformer model, and the Transformer model to extract meteorological features from multivariate meteorological data in different regions; A multi-region load forecasting model is established, including: a multi-scale LSTM feature extraction network composed of multiple LSTM models of different scales, a multi-layer perceptron network, and a multi-region load forecasting network based on a transformer model; wherein the multi-scale LSTM feature extraction network extracts global load features and local load features from load features, and the multi-layer perceptron network fuses the global load features and local load features with meteorological features to obtain an input data set of the multi-region load forecasting network; based on the input data set, the multi-region load forecasting network outputs a multi-region load forecasting result.

2. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 1 is characterized in that: Multivariate meteorological data include: wind speed, wind direction, wind force, humidity, atmospheric pressure, and temperature.

3. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 2 is characterized in that: Process load data from multiple regions and multivariate meteorological data, including: Remove outliers, fill in missing values, and perform time series smoothing on the data set after removing outliers and filling in missing values; Among them, the moving average method is used to perform time series smoothing on the data set after deleting outliers and filling missing values.

4. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 1 is characterized in that: The load feature extraction model includes: input layer, N spatial attention modules, N long short-term memory network modules, and output layer; The multi-region load data is processed by the first spatial attention module and input into the first long short-term memory network module. The output data of the first long short-term memory network module is spliced ​​with the multi-region load data and input into the second spatial attention module. The data processed by the second spatial attention module and the output data of the first long short-term memory network module are input into the second long short-term memory network module together. By analogy, the output data of the N-1th long short-term memory network module is spliced ​​with the multi-region load data and input into the Nth spatial attention module. The data processed by the Nth spatial attention module and the output data of the N-1th long short-term memory network module are input into the Nth long short-term memory network module together. The data output by the Nth long short-term memory network module is the multi-region load feature. The multi-region load feature output by the Nth long short-term memory network module is transmitted to the output layer.

5. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 4 is characterized in that: The structure of each spatial attention module is the same, including: load data input unit, linear mapping layer, activation function unit; The multi-regional load data are input into the linear mapping layer after passing through the load data input unit. The linear mapping layer includes three linear mapping units. The multi-regional load data are respectively obtained through the three linear mapping units to obtain three weighted matrices, which satisfy the following relationship: Q=W Q ·X K=W K ·X V=W V ·X Where Q is the first weight matrix W output by the first linear mapping unit Q The weighted sum matrix of the multi-region load data X is defined as the query matrix, and K is the second weight matrix W output by the second linear unit. K and the weighted sum matrix of multi-region load data X, defined as the key matrix, V m The third weight matrix W output by the third linear emitter unit V and the weighted sum matrix of multi-region load data X, defined as the value matrix; The activation function unit Softmax obtains the output data of the mth spatial attention module according to the following relationship: Where Attention(Q, K, V) is the output data of the spatial attention module, dk is the scaling factor, and the value of the scaling factor dk is the number of columns of the query matrix Q and the key matrix K.

6. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 1 is characterized in that: The meteorological feature extraction model includes: an input layer, N T2T modules, N VisionTransformer modules, N Transformer modules, and an output layer; After passing through the first T2T module, the multivariate meteorological data is first embedded in a patch and then input into the first Vision Transformer module. The data output by the first Vision Transformer module is transmitted to the first Transformer module. After passing through the second T2T module together with the multivariate meteorological data, the data output by the first Transformer module is first embedded in a patch and then input into the second Vision Transformer module. The data output by the second Vision Transformer module and the data output by the first Transformer module are transmitted to the second Transformer module together. Similarly, after passing through the Nth T2T module together with the multivariate meteorological data, the data output by the N-1th Transformer module is first embedded in a patch and then input into the Nth Vision Transformer module. The data output by the Nth Vision Transformer module and the data output by the N-1th Transformer module are transmitted to the Nth Transformer module together. The meteorological features output by the Nth Transformer module TransformerblockN are transmitted to the output layer.

7. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 6 is characterized in that: Each T2T module includes: f processing units and f-1 conversion units; In the conversion unit, the i-th element in the sequence is updated using the weighted sum of the i-th element in the sequence and other elements in the sequence, and a new sequence is obtained after all elements are updated; Among them, the weights used in weighted summation satisfy the following relationship: α[i,j]=exp(score[i,j]) / Sum_k(exp(score[i,j])) Where α[i,j] is the weight between the i-th element and the j-th element, socre[i,j] is the correlation between the i-th element and the j-th element, and Sum_k() is the sum.

8. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 1 is characterized in that: The multi-scale LSTM feature extraction network extracts global load features and local load features from load features, including: Process the load characteristics and meteorological characteristics into a confusion matrix form; The multi-scale LSTM feature extraction network includes N LSTM models of different scales; the first LSTM model takes the load feature in the form of a confusion matrix as input and outputs the first local feature of the load; the second LSTM model takes the first local feature of the load as input and outputs the second local feature of the load; and so on, the Nth LSTM model takes the N-1th local feature of the load as input and outputs the Nth local feature and the global feature of the load.

9. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 8, characterized in that: The multi-layer perceptron network fuses the global and local load features with the meteorological features to obtain the input data set of the multi-region load forecasting network, including: The multi-layer perceptron network includes N multi-layer perceptron layers; In the first multi-layer perceptron layer, the first local feature of the load is fused to obtain the first local fusion feature, and the first local fusion feature matches the dimension of the meteorological feature in the form of a confusion matrix; in the second multi-layer perceptron layer, the second local feature of the load is fused to obtain the second local fusion feature, and the second local fusion feature matches the dimension of the meteorological feature in the form of a confusion matrix; and so on, in the Nth multi-layer perceptron layer, the Nth local feature of the load is fused to obtain the Nth local fusion feature, and the global feature of the load is fused to obtain the global fusion feature, and the Nth local fusion feature and the global fusion feature both match the dimension of the meteorological feature in the form of a confusion matrix; the first local fusion feature, the second local fusion feature, ..., the Nth local fusion feature and the global fusion feature constitute the input data set of the multi-regional load forecasting network.

10. The multi-region load forecasting method integrating global enhancement and local attention features according to claim 9 is characterized in that: Based on the input data set, the multi-region load forecasting network outputs multi-region load forecasting results, including: The multi-region load forecasting network includes N transformer models. The meteorological feature input data is transmitted to the first transformer model together with the first local fusion feature. The output data of the first transformer model is transmitted to the second transformer model together with the second local fusion feature. And so on. The output data of the N-1th transformer model is transmitted to the Nth transformer model together with the Nth local fusion feature and the global fusion feature. The Nth transformer model outputs the multi-region load forecasting result.

11. A multi-region load forecasting system integrating global enhancement and local attention features, characterized in that: include: Data processing module, used to obtain load data and multivariate meteorological data from multiple regions; Based on the long short-term memory network and spatial attention mechanism, a load feature extraction model is established to extract load features from load data in different regions; A meteorological feature extraction model is established based on the token to token model, the Vision Transformer model, and the Transformer model to extract meteorological features from multivariate meteorological data in different regions; The load forecasting module is used to establish a multi-region load forecasting model, including: a multi-scale LSTM feature extraction network composed of multiple LSTM models of different scales, a multi-layer perceptron network, and a multi-region load forecasting network established based on a transformer model; wherein the multi-scale LSTM feature extraction network extracts global load features and local load features from load features, and the multi-layer perceptron network fuses the global load features and local load features with meteorological features to obtain an input data set for the multi-region load forecasting network; based on the input data set, the multi-region load forecasting network outputs multi-region load forecasting results.

12. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Images